Missing value where TRUE/FALSE needed in R

Question

[huynguyen96.dnu]

First, I pre-process my dataset (TCGA breast cancer including GE and CNA) like the way the below paper did on TCGA cutaneous melanoma: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5559859/pdf/12864_2017_Article_3990.pdf (Data analysis, Page 9 of 12)

At the end of pre-processing, I have TCGA BRCA Gene expression (478 genes - 981 samples) and TCGA BRCA CNA (589 genes - 981 samples), assigned as c_exp and c_cna , respectively. For the above study, they process their dataset and then perform integrative analysis using the analysis tool ANCut.

Next, the above group of authors developed a novel tool: AWNCut: https://www.ncbi.nlm.nih.gov/pubmed/30094873 and continue to use the above pre-processed dataset (ANCut) to perform analyses using AWNCut.

Now, I've been applying my pre-processed dataset to AWNCut.

#library
source("AWNCut_fun.R") #https://github.com/shuanggema/AWNCut/blob/master/AWNCut_fun.R
#This sets up the initial parameters
lambda <- seq(1,15,0.5) #Tuning parameter lambda
Tau <- seq(0.1,1.5,0.1) #Tuning parameter tau 
K=7; #Number of clusters 
X=t(c_exp) #row=patients, column=gene
Z=t(c_cna) #row=patients, column=gene

Tune1 <- AWNcut.TuningSelection(X, Z, K, lambda, Tau, B=500, L=1000)

I run into a problem:

Error in if (OP.value <= OP.value.old) {:

Missing value where TRUE/FALSE needed

In addition: there were 50 or more warnings

I do some researches and know that I get this problem when I am trying to compare the missing value (NA, NAN,...) with boolen value (TRUE/FALSE). But when I check my dataset:

tableis.na(c_exp))
#FALSE 
#468918 
tableis.na(c_cna))
#FALSE 
#577809 
table(is.finite(c_exp))
#TRUE 
#468918 
table(is.finite(c_cna))
#TRUE 
#577809

Please help me clarify what problem I am facing? Any suggestion is appreciated!

[russhh]

Have you ran debug(AWNcut) and caledl AWNcut.TuningSelection with your input? From the source code https://github.com/shuanggema/AWNCut/blob/1af5af8498acfdddb69938bab9cbf58fe04de96c/AWNCut_fun.R there seems to be a few places where a missing value could arise

[huynguyen96.dnu]

Hi Russhh, I've just written additionally a code line like you said: debug(AWNcut) and it doesn't report any problem?

And what does "caledl AWNcut.TuningSelection with your input" means? can you figure it out?

[russhh]

Sorry, that was a typo. Call AWNcut.TuningSelection with your input ...

[russhh]

There's an introduction to using the debugger in "Advanced R": https://adv-r.hadley.nz/debugging.html#browser

[huynguyen96.dnu]

> debug(AWNcut)
> Tune1 <- AWNcut.TuningSelection(X, Z, K, lambda, Tau, B=500, L=1000)
debugging in: AWNcut(X, Z, K, lambda, Tau, B, L = 1000)
debug at AWNCut_fun.R#56: {
    X <- scale(X)
    Z <- scale(Z)
    Para <- as.data.frame(cbind(rep(lambda, each = length(Tau)), 
        rep(Tau, length(lambda))))
    out <- list()
    for (para in 1:nrow(Para)) {
        lam <- Para[para, 1]
        tau <- Para[para, 2]
        p1 <- ncol(X)
        p2 <- ncol(Z)
        w1 <- rep(1/sqrt(p1), p1)
        w2 <- rep(1/sqrt(p2), p2)
        b <- 0
        ws.old <- c(w1, w2)
        ws <- rep(0, p1 + p2)
        Cs.old <- matrix(rep(0, nrow(Z) * K), nrow(Z), K)
        for (i in 1:nrow(Z)) {
            Cs.old[i, sample(K, 1)] <- 1
        }
        while ((b <= B) || (sum(ws - ws.old)/sum(ws.old) >= 0.001)) {
            b <- b + 1
            wm1 <- AWNcut.W(X, Z, ws.old)
            WX1 <- wm1[[1]]
            WZ1 <- wm1[[2]]
            a1 <- AWNcut.OP(X, Z, WX1, WZ1, Cs.old, tau)
            OP.value.old <- a1$TOP + lam * sum(ws.old * a1$Cor.perfeature)/(p1 + 
                p2)
            Cs <- AWNcut.UpdateCs(WX1, WZ1, K, Cs.old)
            ws <- AWNcut.UpdateWs(X, Z, K, WX1, WZ1, b, Cs, ws.old, 
                tau)
            wm2 <- AWNcut.W(X, Z, ws)
            WX2 <- wm2[[1]]
            WZ2 <- wm2[[2]]
            a2 <- AWNcut.OP(X, Z, WX2, WZ2, Cs, tau)
            OP.value <- a2$TOP + lam * sum(ws * a2$Cor.perfeature)/(p1 + 
                p2)
            if (OP.value <= OP.value.old) {
                des <- rbinom(1, 1, Prob(OP.value, OP.value.old, 
                  L, b))
                if (des == 1) {
                  Cs.old <- Cs
                  ws.old <- ws
                }
                else {
                  Cs <- Cs.old
                  ws <- ws.old
                }
            }
            else {
                Cs.old <- Cs
                ws.old <- ws
            }
        }
        out[[para]] <- list(lambda = lam, tau = tau, Cs = Cs.old, 
            ws = ws.old, OP.value = OP.value)
    }
    return(out)
}
Browse[2]> AWNcut.TuningSelection
function(X, Z, K, lambda, Tau, B=500, L=1000){
  out <- AWNcut(X, Z, K, lambda, Tau, B, L=1000)
  Para <- as.data.frame(cbind(rep(lambda,each=length(Tau)),rep(Tau,length(lambda))))
  dbi <- NULL
  for(i in 1:nrow(Para)){
    Cs <- out[[i]]$Cs
    ws <- out[[i]]$ws
    dbi <- c(dbi, DBI(cbind(X,Z),K,Cs,ws))
  }
  return(list(num=which.max(dbi),Table=t(cbind(Para,dbi)), lam=Para[which.max(dbi),1], tau=Para[which.max(dbi),2], DBI=max(dbi)))
}
<bytecode: 0x1409c5710>

[huynguyen96.dnu]

#This sets up the initial parameters
lambda <- seq(1,15,0.5) #Tuning parameter lambda
Tau <- seq(0.1,1.5,0.1) #Tuning parameter tau 
K=7; #Number of clusters 
X=t(c_exp) #row=patients, column=gene
Z=t(c_cna) #row=patients, column=gene

Tune1 <- AWNcut.TuningSelection(X, Z, K, lambda, Tau, B=500, L=1000)

I assigned c_exp to X, c_cna to Z, K=7, lambda = seq(1,15,0.5), and Tau <- seq(0.1,1.5,0.1) as you can see, in which c_exp and c_cna are my two data as matrix equivalent to gene expression and CNA, respectively